File size: 2,580 Bytes
8d19768
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
---
license: mit
library_name: "trl"
tags:
- DPO
- WeniGPT
base_model: Weni/WeniGPT-2.2.3-Zephyr-7B-merged-LLM_Base_2.0.3_SFT
model-index:
- name: Weni/WeniGPT-2.8.1-Zephyr-7B-zephyr-prompt-DPO-binarized
  results: []
language: ['en']
---

# Weni/WeniGPT-2.8.1-Zephyr-7B-zephyr-prompt-DPO-binarized

This model is a fine-tuned version of [Weni/WeniGPT-2.2.3-Zephyr-7B-merged-LLM_Base_2.0.3_SFT] on the dataset HuggingFaceH4/ultrafeedback_binarized with the DPO trainer. It is part of the WeniGPT project for [Weni](https://weni.ai/).

It achieves the following results on the evaluation set:
{'eval_loss': 1.6283894777297974, 'eval_runtime': 99.8179, 'eval_samples_per_second': 2.004, 'eval_steps_per_second': 0.501, 'eval_rewards/chosen': 0.8732965588569641, 'eval_rewards/rejected': -6.086017608642578, 'eval_rewards/accuracies': 0.5649999976158142, 'eval_rewards/margins': 6.959313869476318, 'eval_logps/rejected': -357.7032165527344, 'eval_logps/chosen': -362.0609130859375, 'eval_logits/rejected': -2.5749547481536865, 'eval_logits/chosen': -2.5779407024383545, 'epoch': 1.0}

## Intended uses & limitations

This model has not been trained to avoid specific intructions. 

## Training procedure

Finetuning was done on the model Weni/WeniGPT-2.2.3-Zephyr-7B-merged-LLM_Base_2.0.3_SFT with the following prompt:

```
Prompt:
<|user|>{prompt}</s>


Chosen:
<|assistant|>{chosen}</s>


Rejected:
<|assistant|>{rejected}</s>
```

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.0002
- per_device_train_batch_size: 4
- per_device_eval_batch_size: 4
- gradient_accumulation_steps: 4
- num_gpus: 1
- total_train_batch_size: 16
- optimizer: AdamW
- lr_scheduler_type: cosine
- num_steps: 112
- quantization_type: bitsandbytes
- LoRA: ("\n  - bits: 4\n  - use_exllama: True\n  - device_map: auto\n  - use_cache: False\n  - lora_r: 16\n  - lora_alpha: 16\n  - lora_dropout: 0.05\n  - bias: none\n  - target_modules: ['q_proj', 'k_proj', 'v_proj', 'o_proj']\n  - task_type: CAUSAL_LM",)

### Training results

### Framework versions

- transformers==4.38.2
- datasets==2.17.1
- peft==0.8.2
- safetensors==0.4.2
- evaluate==0.4.1
- bitsandbytes==0.42
- huggingface_hub==0.20.3
- seqeval==1.2.2
- optimum==1.17.1
- auto-gptq==0.7.0
- gpustat==1.1.1
- deepspeed==0.13.2
- wandb==0.16.3
- trl==0.7.11
- accelerate==0.27.2
- coloredlogs==15.0.1
- traitlets==5.14.1
- autoawq@https://github.com/casper-hansen/AutoAWQ/releases/download/v0.2.0/autoawq-0.2.0+cu118-cp310-cp310-linux_x86_64.whl

### Hardware
- Cloud provided: runpod.io